Just Walk Out debuted to the public in early 2018 at an Amazon Go store in Seattle. The original version was built using generative AI and the leading-edge machine learning available at the time, relying on cameras and sensors to track what shoppers removed from shelves and what they put back. However, Amazon stated, in unusual or novel shopping scenarios (such as if a camera view was obscured), the sequential approach could take time to determine the correct purchases, which could then trigger a manual retraining of the model.
To account for these scenarios, the new multimodal foundation model increases accuracy by analyzing all sensor data simultaneously, rather than sequentially, and prioritizes what’s most important to determine the variety and quantity of items selected and lessening the chance of an inaccurate count or item getting recorded.
“The improvements to our AI system are so seamless that you will continue to enjoy the same contactless checkout-free shopping experience you’ve come to expect at Just Walk Out stores, all while protecting your privacy,” said Jon Jenkins, vice president, Just Walk Out technology, AWS applications.
Amazon also believes the new model will reduce the need for model retraining in unfamiliar shopping scenarios, as the self-learning system will continue to teach itself.
Just Walk Out is currently available in more than 170 third-party locations throughout the United States, the United Kingdom, Australia and Canada. The company plans to launch additional stores in 2024, with the goal of more than doubling the number of locations with the technology by the end of the year.